Before handing an AI project to a freelancer, ask these 7 concrete questions about scope, data, real costs, code ownership and maintenance. A buyer's guide for founders and managers, without jargon or magic promises.
Most AI projects do not fail because the model was weak. They fail because the scope was vague, the data was not ready, or nobody asked who would maintain the thing six months later. If you are hiring a freelance AI consultant, the quality of your questions determines the quality of your outcome. Here are seven questions worth asking before you sign anything.
1. Can you explain what you would build without using the word "AI"?
A good consultant can describe your project in plain business terms: "a system that reads incoming supplier emails and fills your purchase spreadsheet automatically." If every answer circles back to model names and frameworks, you are talking to someone selling technology rather than solving a problem. The first deliverable of any serious engagement is a clear problem statement, not an architecture diagram.
2. What exactly is in scope, and what is explicitly out?
Ask for a written scope with three lists: what will be delivered, what is assumed to exist already, and what is excluded. Watch for assumptions about your side: clean data, API access, a test environment, someone available to review outputs. These are where budgets quietly double. A precise quote on a narrow scope beats a vague quote on everything.
3. What does my data actually look like today, and what will it take to use it?
This is the question that separates experienced consultants from enthusiastic ones. Ask how they will audit your data before promising results. A serious answer includes: where the data lives, its format and quality, whether you have the legal right to use it, and what happens to sensitive information. For most small and mid-sized companies, data preparation is 50 to 70 percent of the effort. If that is not in the plan, the plan is fiction.
4. How will we measure success, and what happens if it is not reached?
Define one or two concrete metrics before starting. Not "improve efficiency" but "reduce invoice processing time from 8 minutes to under 2" or "route 80 percent of support tickets correctly." Then agree on what happens if the target is missed: a revised approach, a partial refund, or a clearly bounded pilot that simply ends. Pilots with a fixed budget and a decision date are far healthier than open-ended commitments.
5. What are the real costs beyond your fee?
The consultant's rate is rarely the whole bill. Ask directly about: API usage fees from providers such as Claude, OpenAI or DeepSeek, hosting, vector databases, monitoring tools, and the internal time your team must contribute. Request a rough monthly running cost at your expected volume. An honest consultant will give you a range and explain what drives it up.
6. Who owns the code, the prompts and the data, and where do they live?
Get this in writing. You should own the application code and your data; the consultant may retain rights to generic internal libraries, which is reasonable. Ask where everything is hosted and whether you can move it. Avoid systems locked inside a proprietary platform you cannot export from. Also ask for documentation and a handover session, so you are not permanently dependent on one person.
7. What happens after launch?
AI systems drift. Inputs change, providers update models, edge cases appear. Ask what monitoring is included, how failures are detected, and what a maintenance arrangement looks like. A monthly retainer for a few hours of monitoring and adjustment is normal. "It will just work" is not an answer.
Two extra signals worth noticing
First, does the consultant ask about your users and your existing workflow before proposing tools? The best implementations fit how people already work. Second, do they mention what they would *not* use AI for? A specialist who tells you a simple rules-based script or a better spreadsheet would solve your problem faster is saving you money, not losing a sale.
A quick scoring approach
Rate each answer from 1 to 3 and total it. Below 12, keep looking. Between 12 and 17, proceed but tighten the contract. Above 17, you have found someone who thinks like a partner. When comparing quotes, normalize them: same scope, same assumptions, same timeline, same ownership terms. The cheapest proposal is often the one with the most undefined work hiding inside it.
Let's talk about your project
If you are evaluating an AI initiative and want a second opinion on scope, data readiness or realistic costs, I am happy to look at what you have. No pitch, no jargon, just a practical conversation about whether AI is the right answer and what a sensible first step would look like. You can reach me at contact@hamzabelgacem.com.